Google Search MCP: Live, Rich Google Results for AI

The SerpSearch Google Search MCP is now available at https://mcp.serpsearch.com/mcp. It gives Codex, Claude Code, and other compatible AI clients direct access to live Google Search data through the Model Context Protocol.

Connect it once, approve access through OAuth, and the assistant can choose the right SerpSearch tool for the question in front of it. That includes ordinary web results, rich Google result modules, Maps places, place details, reviews, products, news, images, videos, translations, autocomplete suggestions, and asynchronous review exports.

There are 15 tools in the initial release. A normal tool call consumes the same single API request as its matching REST endpoint. MCP discovery, OAuth, tool listing, and review batch status checks don't consume requests.

What is a Google Search MCP?

An MCP server exposes tools that an AI client can discover and call during a conversation. Instead of copying search results into a prompt or writing a separate integration, you can ask the assistant a question that needs current Google data. The assistant sends structured arguments to SerpSearch and receives structured results it can reason over.

You may also see this described as a Google Scrape MCP. The practical goal is the same: give the model structured Google data without making it parse result pages or maintain a scraping stack itself.

For example, you could ask:

• “Compare the current stock price, recent news, and major Google Search results for Apple and Microsoft.”
• “Find well-rated coffee shops near Union Square that are open now, then compare their review themes.”
• “Show me iPhone offers with their displayed prices, ratings, merchants, and product links.”
• “Export every review for this Place ID from January through June and tell me when the file is ready.”

This is useful when the answer depends on information that may have changed since the model was trained. The model handles the conversation and analysis. SerpSearch supplies the current search data.

Rich Google Search results, not just ten blue links

The google_search tool returns organic results and the rich modules Google displays for the query. Depending on the search, that can include:

• AI Overviews with answer text, structured steps, citations, and products
• knowledge graphs and entity facts
• local packs and business details
• products with prices, merchants, ratings, discounts, availability, delivery terms, variants, and links
• top stories, jobs, hotels, flights, sports, movies, images, and related searches
• calculators, translations, currency conversion, and market data

Responses are intentionally sparse. If Google doesn't display a product module, AI Overview, local pack, or another feature for a query, that field is omitted instead of being filled with invented or empty data. Your assistant can inspect what was actually returned and adapt its answer.

That distinction is what makes this a Rich SERP MCP rather than a basic link lookup. It can answer with the structure Google exposes, including sources and URLs that a person can inspect afterward. If you're looking for a Rich Google Search Results MCP, the complete response schema is documented in the Google Search API reference.

Google Products MCP capabilities

Product research often needs more than an organic result title. When Google displays product cards or shopping offers, SerpSearch can return the visible product name, current price, merchant, rating, review count, discount, availability, delivery information, variants, thumbnail, and destination link.

That lets an assistant compare offers without guessing which nearby text belongs to which product. It can preserve merchant and source information in the answer, or hand the structured records to another step in an agent workflow.

Calling this a Google Products MCP doesn't mean every search is forced into a shopping result. Product fields appear when Google returns that module for the selected query, country, and language. Broader queries may return organic results, news, a knowledge graph, or a mixture of modules instead.

Google Reviews MCP and local research

The MCP includes separate tools for Google Maps search, place details, driving distance, and reviews. A model can find a business, resolve its Google Place ID or Feature ID, inspect its details, and load paginated reviews without asking the user to translate identifiers by hand.

The standard google_reviews tool is the right fit for interactive review questions or a few pages of recent reviews. Pro, Business, and Enterprise accounts can also use five review batch tools to create, monitor, cancel, list, and download a date-range export.

Batch exports are asynchronous and produce a private gzipped NDJSON file. Completed files remain available for seven days, and each signed download URL is valid for 15 minutes. Failed or cancelled exports consume zero requests. If the account's available quota is reached first, the job can deliver the partial file and charge only the page requests used.

Our 10,528-review production benchmark explains when batch exports make sense, how their usage is calculated, and how they compare with sequential review pagination.

Other tools included at launch

The server currently exposes these 15 tools:

Together, they cover the common AI Google Search workflow: find current information, inspect the sources, follow local or product-specific data when it exists, and return a useful answer in the same conversation.

Connect SerpSearch to Codex

Add the remote HTTP server with one command:

codex mcp add serpsearch --url https://mcp.serpsearch.com/mcp

On its first connection, Codex will open the SerpSearch authorization flow. Sign in, review the requested permissions, and approve the connection. You don't need to paste an API key into the client.

For clients that use JSON configuration, add:

{
  "mcpServers": {
    "serpsearch": {
      "type": "http",
      "url": "https://mcp.serpsearch.com/mcp"
    }
  }
}

The endpoint uses OAuth 2.1 with PKCE. Standard search tools request the mcp:tools permission, while review batch tools also require mcp:batches. A verified SerpSearch account is required to approve access.

See the MCP setup reference for configuration, authentication, scopes, tool descriptions, billing behavior, and retry guidance.

Usage and cost expectations

A request made through MCP is priced and counted the same way as the equivalent SerpSearch REST request. There isn't a separate MCP surcharge. Your existing plan quota, requests-per-second limit, usage history, and automatic top-up settings still apply.

This matters for agent workflows because a conversational request isn't always the same as one tool call. A simple factual lookup may need one search. A comparison could use several searches plus place details or reviews. The client decides which tools to call, so review its tool activity when request count needs to be predictable.

Tool calls accept an optional idempotency_key for safe retries. Reusing the same key after an interrupted call returns the stored result without consuming quota twice. Batch creation requires an idempotency key because creating the same long-running export twice would waste time and quota.

For large review exports, page requests are reserved while the job is running and charged together only after a full or quota-limited partial file is delivered. Creating the job, checking its status, and requesting its download link don't consume monthly request quota, though those operations still count against the account's rate limit.

Current quotas, rate limits, and top-up pricing are listed on the pricing page.

When to use the MCP or the REST API

Use the MCP when a model needs to decide what to search, interpret rich results, call a follow-up tool, or explain current information inside a conversation. It removes the application code normally needed to translate model output into API requests and feed each response back into the model.

Use the REST API when your software already knows the exact endpoint and parameters it needs, or when deterministic request orchestration matters more than conversational tool selection. Both interfaces reach the same SerpSearch data and share the same account usage.

The MCP is available now. Create a SerpSearch account, connect https://mcp.serpsearch.com/mcp, and ask your client a question that requires current Google data. The full list of tools and their permissions is in the MCP documentation.